It’s a tool. There are uses. However we’re currently in the ‘if all you have is a hammer, then everything is a nail’ phase. Overhyped, oversold.

Whats often left out is that it’s often trained illegally and should carry the same punishment that private people have to face on illegal acquisition of things.

Basically it doesn’t matter what it is. It’s the current thing that’s financially blown way out of proportion to shovel our money around and to a select few, if we want it or not.

I’m usually not against “tools” but I have a hard time finding something useful with LLMs.

The only good usage of LLMs is to optimize algorithms in complex code. Something like “move this variable and your application will be 100x faster.” But even for that case, we could have alternatives that do not involve LLMs.

That’s it, everything else is bad and useless:

• mega companies profiting from piracy without being sued
• AI companies who publicly declare that they want us to give up our brains and become slaves
• layoffs (it’s never officially because of AI, but it’s always a manager thinking that he can fire people or not hire new guys because Cursor is good enough for the peasants)
• employees have way more work without being compensated
• making people stupid
• making students more lazy than they are, and allowing them to cheat (cheating required some dedication, not it’s free for all)
• artists in all fields are being fucked and may not live from their art
• replacing customer service with chatbots that prevent people from getting answers
• bots everywhere
• spamming forums (look at what happens right now on Lemmy)
• DDoS and attacks on web sites
• enabling script kiddies to shit on the internet
• hallucination and lies
• actually, LLM are 100% hallucinating all the time, the fact that they are coherent is a statistical side-effect
• data centers, pollution, pissing people off when they live around one data center
• generated images, it’s really creepy for some reason (even for a simple restaurant menu)

The examples I gave are real and I have experienced them first hand, it’s not an anti-AI crusade. For the first time in my life, I don’t see how a technology can be useful.

I use it for coding, for translation and for some irl chore / bureaucracy. It’s very useful to me, though not without flaws and dealing with some of its stupidity. I am now finally in a position where if I can’t find a good existing program that solves my problem - I can whip up exactly what I want in a week (as opposed to a 6 month endeavour)

I don’t use image/video/audio generation

Shit peaked before 9/11 when SmarterChild dropped on AIM

I think it’s a tool, but a tool that has been weaponized. I think “AI” (I use quotes because we are really talking about LLMs primarily) have some fantastic applications for automation, things like dictation, pattern recognition, etc.

But unlike non-AI tools that can do similar tasks (linters, non ai machine translation, dictation, hell even search engines at this point) AI does the task with the confidence of an expert, speaking in a confident voice, to people who generally don’t know if the information is accurate. It’s being sold as a knowledge hub, when in reality it is a tool, one that can be wrong and does not think. And the marketing is agressive. They want you to forget how to think for yourself so you have to pay a subscription for thinking.

AI psychosis is real, I’ve seen it happen in real time with my coworkers. The quality of our new hires has gone downhill because none of them seem to know how to solve a problem. I often hear “Claude thinks this” in response to issues that are just barely above basic level. There is no longer a desire to find knowledge or puzzle out a problem when it can be cognitively offloaded to the llm. There is danger in the very service that the llm provides - reduction in friction in your daily life also equates to a reduction in learning. Which sucks because noodling out complex problems with an llm as a rubber duck can actually be a great learning experience in situations where a product expert isn’t around to bounce off of. But it comes with the caveat of knowing how to check the answers and having the curiosity to delve into those answers and dig through them yourself. Without that crucial step you are just offloading your brain to the computer to think for you and you learn nothing.

And I can’t blame the AI psychosis on stupid people. It’s being marketed as an all knowing infallible assistant. Yet tangibly many every day people find their lives made worse for it. AI has been shoved into everything when it isn’t even as good as the thing it is replacing, or it fails more often.

Then we get to the actual generative functions of ai. Generative art, be it picture or text, is trained without permission or compensation, on the work of millions of artists, with the goal of making those artists obsolete. As an artist I’ve seen every single artistic industry be hit with this, from stock photography to furry porn to high budget movies. And much like the previous use cases of AI replacing currently existing software, it’s replacing artists with less quality products. AI art is recognizable because it is homogenized by default. And sure it’s tweakable, but by default it is recognizable as this uncanny homogenized “good enough” picture to get an idea across. And much like everything else it’s being used as a way to capitalize on industries, because at the end of the day the only reason to use generative ai for art is to make it faster and cheaper. Every single person that uses it that I’ve seen has used it for the same reasons. “I don’t have time to learn x” or “I need an idea out now” or “I can make more by doing this”. Its because art has become completely monitized. People don’t create for the sake of creation now, it’s for a job, or for a deadline, or for clout. Artists who love their craft and are not bogged by deadlines or bills are overwhelming against AI in general, as artists are some of the most vocal anti generative AI crowd.

When we look at all of this, AI is a tool that has some great localized use cases, especially in pattern recognition, dictation, language, etc. but it’s been coopted as a brain subscription by the people in charge of training it and selling it to us. It’s been shoehorned into every aspect of our lives, made everything more expensive, made people dumber, collectively destroyed the lives of artists off the backs of their work, and enshittified almost every single industry.

The tradeoffs aren’t worth it.

Edit: also forgot to add the privacy nightmare that comes from AI surveillance

Mixed feelings. I’m kind of an AI pilled doomer.

I see that AI can be a productivity boost - if used correctly in the right places. But I’m also increasingly worried that it might one day replace any work that’s done on a computer (which is most white collar work).

I’m doing a lot of data processing work. I’ve been able to see AI do all this in one go:

• Read a vague Jira task item about some data error.
• Diagnose on real data what the underlying problem might be.
• Find the line of code that need to be changed.
• Rerun the data pipeline with the new line of code.
• Verify on new data that the error has been fixed.
• Submit a pull request describing the change with a detailed report on impact analysis.

Is it perfect? No, it’s often making mistakes and digs deep into the wrong rabbit holes. But is it crazy that it has managed to do all of this once? Yes it is. This was unimaginable just a few years ago.

If this level of automation going to continue in this pace, what’s left for us humans?

Too broad of an question to answer to.

I use it quite a bit. It replaced search engines for me and I like to have conversations about esoteric topics that no other people in my life is interested in talking about. It’s by far the most fascinating piece of new technology that has been invented during my lifetime.

But I also think it’ll lead to our destruction eventually. Not GenAI but AGI/ASI.

As long as I can still detect it I block those who use it.

I’m conflicted with a lot of it. I hate seeing ai generated content when I’m not looking for it but there are also some games that are pretty fun that rely on LLMs to work. Origins was a pretty good proof of concept and things like fantasy parliament would be amazing if they weren’t so resource intensive and could be run locally.

My main problem is just how inefficient that side of it is, it shouldn’t use the same computing power to run your own local model as it does to run something as insanely detailed as RDR2 or cyberpunk 2077.

I have a dirt cheap VPS running about 30 or so services all made by AI. They automate lots of meanial tasks and help me save money.

Tasks I would never care to do, can now be done for me. It’s a suboptimizers wet dream. The cost of personally tailored software is now really low.

Since I want to write in a bit more detail about my personal LLM/AI journey so far I will just give a Tl;Dr here: I started as a hater for that barely functional piece of junk that dared to call itself AI and slowly started to come around once it became better and more reliable. Now, let’s start with the story time.

I still remember when cleverbot was the best way to talk to a computer and it was pretty shit, then suddenly ChatGPT popped up one day. I obv. had to give it a try and I was amazed that we already got that far and it happened basically over night (I know now that at that point the first LLMs existed for quite some time but it took a while for them to blow up and therefore me discovering their existence). At that point, they were pretty inacurate, couldn’t even do simple math (which I found really ironic since that is basically the only thing a computer can do and very well too) and was inaccurate most of the time. I always had to double check each output and had to manually fix it pretty much every time, if it even was salvagable in the first place. But usually, I was better and faster at any task I could possibly give it so I became anti AI, started hating on the technogy and mocking people who used it.

Half a year later I finally decided to give it another try and it actually improved. Simple math was correct most of the time, it started to get most facts straight and some tasks like formatting text was either correct or just had some small mistakes I could easily fix myself. That was the turning point for me where I stated using it more often and hating it less.

Now, I actually like AI/LLMs. I use it almost daily for questions about pretty much anything that isn’t about recent news (which is still very prone to errors) or about a nieche topic (same issue). The main reason why is because I stongly prefer getting taught new things with the chance to ask follow up questions or to clarify things I didn’t underatand. I also use it to do simple tasks for me I don’t want to do myself, like formatting texts in specific ways. And I also use image gen whenever I need an image, be it for my profile picture, banners or anything else, since I hate drawing stuff myself, do not have the money to commission stuff from actual artists and noone but AIs want to do it for free.

And as a developer, I of couse use AI coding tools. However, I have to say that I’m not a fan of agentic coding, better known as vibe coding to most non-developers, where you tell the AI what you want and it does all the coding for you. In my experience, the result sucks. Too many unneccesary comments, too many checks that usually just degrade the performance of the software and at some point, shit just starts breaking, the AI can’t fix it and the codebase is too messy for me to read into it and fix it myself so I have to start over and do it myself anyway. For small stuff, like a simple script to inject into a website to make a table sortable or a cronjob that deletes every file in a temp folder that is older than a week, I have no issue with letting an AI vibecode that for me, then reading through the result to make sure it actually does what it should before using it. But for larger projects, I just give the AI access to my codebase and then chat with it when I’m stuck, unsure which approach fits best or give it smaller tasks to do for me and always check the output manually, never trust it fully. So after auto-completing words, then lines, then code snippets/chunks, I see coding LLMs as the next step that lets me generate even lager chunks, or even entire files, but with a more cautious approach. And it replaced the rubber duck for debugging.

One thing I almost never do is use AI to generate, enhance or even correct texts I’m writing. Take this comment for example, I bet I’ve made multiple mistakes and an AI could find better words and write everything way shorter while still getting the important bits right, but for some reason I do prefer to rather keep the human imperfections in it. Only for more… professional settings, I sometimes rely on AI to do a bettrr job at writing than me, and most of the time, it delivers.

Now, what I am still concerned about is the future of AI. What I haven’t mentioned so far is that I never paid for AIs and never will. If they are trained on stolen works where the creators of said works never received any compensation, neither will any AI company see anything from me. But as long as I can use it for free, I will continue doing so. However, I fear that this will soon change, since investor funding will not last forever and looking at the current hardware market and electricity costs, they will need a whole lot of money to keep going. And I’m pretty sure ads won’t cut it this time so I’m really curious how this will end: Will the bubble burst and every large AI company will have to downscale massively to survive or will enough people decide to pay to keep them running the way they do right now? Once again, I will enjoy the benefits while they last and run local models once they try to start charging me, or just return to doing everything myself like before, but I will never support them financially.

Started like you. Fascinated by the reponses i was getting from the LLMs. Then started to notice hallucinations, read.up a bit, and deduced its a fundamental flaw of the approach they use. Its not.going to ‘get better’, because as i said, its fundamental to the methodology.

Nowadays, its a nice conversation piece, or a natural language interface to search. Not worth the investment.

Given my background in CS though, i still believe in traditional ai methods and wish the funding went into that instead. Some specialized ais, e.g. medical imaging are pretty good. Others are glorified brute force or hueristic filtering of large data sets. Not those.

it has rotten peoples brains to the point they cant function without asking it to hallucinate more.

It’s a tech.

Do I hate JavaScript for what it has done to the web? No, I hate people who use JavaScript in their webpages with evil intent.

I have a mentality and way of seeing things that doesn’t really allow me to hate a technology.

That being said, I really dislike all the AI content that people put online. Searching anything has become a pain. And the amount of webpages that are AI written is outrageous.

when it first came out I was skeptical of their claims they wouldn’t replace people with it, because what company wouldn’t replace a person with a bot.

as it became apparent that not only was my original mistrust of them appropriate, but that they were literally breaking laws to build it, my dislike solidified into loathing.

then they began to force the use of it through companies, selling it as a product that can improve productivity while studies had proven the claims as false. my loathing precipitated to hatred when people started losing their jobs for not using it or just being replaced by it.

finally when my government began to use it to kill and oppress, my feelings about AI changed into those of active rebellion.

now, when faced with interacting with AI at work I perform as an idiot. I pretend I don’t understand how to use the tools, or how to set them up. I pretend that I don’t understand what the slop means that someone just vomited up in chat. I pretend that I am such a useless idiot that the only way to help is by personally interacting with me.

weaponized incompetence is a fine line to walk, but I’ve gotten pretty good at it over the years.

so yeah, fuck AI, fuck LLMs, fuck anyone who willingly supports fascist tools.